1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/ceph/ceph_debug.h> 3 4 #include <linux/module.h> 5 #include <linux/fs.h> 6 #include <linux/slab.h> 7 #include <linux/string.h> 8 #include <linux/uaccess.h> 9 #include <linux/kernel.h> 10 #include <linux/writeback.h> 11 #include <linux/vmalloc.h> 12 #include <linux/xattr.h> 13 #include <linux/posix_acl.h> 14 #include <linux/random.h> 15 #include <linux/sort.h> 16 #include <linux/iversion.h> 17 18 #include "super.h" 19 #include "mds_client.h" 20 #include "cache.h" 21 #include <linux/ceph/decode.h> 22 23 /* 24 * Ceph inode operations 25 * 26 * Implement basic inode helpers (get, alloc) and inode ops (getattr, 27 * setattr, etc.), xattr helpers, and helpers for assimilating 28 * metadata returned by the MDS into our cache. 29 * 30 * Also define helpers for doing asynchronous writeback, invalidation, 31 * and truncation for the benefit of those who can't afford to block 32 * (typically because they are in the message handler path). 33 */ 34 35 static const struct inode_operations ceph_symlink_iops; 36 37 static void ceph_inode_work(struct work_struct *work); 38 39 /* 40 * find or create an inode, given the ceph ino number 41 */ 42 static int ceph_set_ino_cb(struct inode *inode, void *data) 43 { 44 struct ceph_inode_info *ci = ceph_inode(inode); 45 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 46 47 ci->i_vino = *(struct ceph_vino *)data; 48 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino); 49 inode_set_iversion_raw(inode, 0); 50 percpu_counter_inc(&mdsc->metric.total_inodes); 51 52 return 0; 53 } 54 55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino) 56 { 57 struct inode *inode; 58 59 inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare, 60 ceph_set_ino_cb, &vino); 61 if (!inode) 62 return ERR_PTR(-ENOMEM); 63 64 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode), 65 ceph_vinop(inode), inode, !!(inode->i_state & I_NEW)); 66 return inode; 67 } 68 69 /* 70 * get/constuct snapdir inode for a given directory 71 */ 72 struct inode *ceph_get_snapdir(struct inode *parent) 73 { 74 struct ceph_vino vino = { 75 .ino = ceph_ino(parent), 76 .snap = CEPH_SNAPDIR, 77 }; 78 struct inode *inode = ceph_get_inode(parent->i_sb, vino); 79 struct ceph_inode_info *ci = ceph_inode(inode); 80 81 BUG_ON(!S_ISDIR(parent->i_mode)); 82 if (IS_ERR(inode)) 83 return inode; 84 inode->i_mode = parent->i_mode; 85 inode->i_uid = parent->i_uid; 86 inode->i_gid = parent->i_gid; 87 inode->i_mtime = parent->i_mtime; 88 inode->i_ctime = parent->i_ctime; 89 inode->i_atime = parent->i_atime; 90 ci->i_rbytes = 0; 91 ci->i_btime = ceph_inode(parent)->i_btime; 92 93 if (inode->i_state & I_NEW) { 94 inode->i_op = &ceph_snapdir_iops; 95 inode->i_fop = &ceph_snapdir_fops; 96 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */ 97 unlock_new_inode(inode); 98 } 99 100 return inode; 101 } 102 103 const struct inode_operations ceph_file_iops = { 104 .permission = ceph_permission, 105 .setattr = ceph_setattr, 106 .getattr = ceph_getattr, 107 .listxattr = ceph_listxattr, 108 .get_acl = ceph_get_acl, 109 .set_acl = ceph_set_acl, 110 }; 111 112 113 /* 114 * We use a 'frag tree' to keep track of the MDS's directory fragments 115 * for a given inode (usually there is just a single fragment). We 116 * need to know when a child frag is delegated to a new MDS, or when 117 * it is flagged as replicated, so we can direct our requests 118 * accordingly. 119 */ 120 121 /* 122 * find/create a frag in the tree 123 */ 124 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci, 125 u32 f) 126 { 127 struct rb_node **p; 128 struct rb_node *parent = NULL; 129 struct ceph_inode_frag *frag; 130 int c; 131 132 p = &ci->i_fragtree.rb_node; 133 while (*p) { 134 parent = *p; 135 frag = rb_entry(parent, struct ceph_inode_frag, node); 136 c = ceph_frag_compare(f, frag->frag); 137 if (c < 0) 138 p = &(*p)->rb_left; 139 else if (c > 0) 140 p = &(*p)->rb_right; 141 else 142 return frag; 143 } 144 145 frag = kmalloc(sizeof(*frag), GFP_NOFS); 146 if (!frag) 147 return ERR_PTR(-ENOMEM); 148 149 frag->frag = f; 150 frag->split_by = 0; 151 frag->mds = -1; 152 frag->ndist = 0; 153 154 rb_link_node(&frag->node, parent, p); 155 rb_insert_color(&frag->node, &ci->i_fragtree); 156 157 dout("get_or_create_frag added %llx.%llx frag %x\n", 158 ceph_vinop(&ci->vfs_inode), f); 159 return frag; 160 } 161 162 /* 163 * find a specific frag @f 164 */ 165 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f) 166 { 167 struct rb_node *n = ci->i_fragtree.rb_node; 168 169 while (n) { 170 struct ceph_inode_frag *frag = 171 rb_entry(n, struct ceph_inode_frag, node); 172 int c = ceph_frag_compare(f, frag->frag); 173 if (c < 0) 174 n = n->rb_left; 175 else if (c > 0) 176 n = n->rb_right; 177 else 178 return frag; 179 } 180 return NULL; 181 } 182 183 /* 184 * Choose frag containing the given value @v. If @pfrag is 185 * specified, copy the frag delegation info to the caller if 186 * it is present. 187 */ 188 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 189 struct ceph_inode_frag *pfrag, int *found) 190 { 191 u32 t = ceph_frag_make(0, 0); 192 struct ceph_inode_frag *frag; 193 unsigned nway, i; 194 u32 n; 195 196 if (found) 197 *found = 0; 198 199 while (1) { 200 WARN_ON(!ceph_frag_contains_value(t, v)); 201 frag = __ceph_find_frag(ci, t); 202 if (!frag) 203 break; /* t is a leaf */ 204 if (frag->split_by == 0) { 205 if (pfrag) 206 memcpy(pfrag, frag, sizeof(*pfrag)); 207 if (found) 208 *found = 1; 209 break; 210 } 211 212 /* choose child */ 213 nway = 1 << frag->split_by; 214 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t, 215 frag->split_by, nway); 216 for (i = 0; i < nway; i++) { 217 n = ceph_frag_make_child(t, frag->split_by, i); 218 if (ceph_frag_contains_value(n, v)) { 219 t = n; 220 break; 221 } 222 } 223 BUG_ON(i == nway); 224 } 225 dout("choose_frag(%x) = %x\n", v, t); 226 227 return t; 228 } 229 230 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v, 231 struct ceph_inode_frag *pfrag, int *found) 232 { 233 u32 ret; 234 mutex_lock(&ci->i_fragtree_mutex); 235 ret = __ceph_choose_frag(ci, v, pfrag, found); 236 mutex_unlock(&ci->i_fragtree_mutex); 237 return ret; 238 } 239 240 /* 241 * Process dirfrag (delegation) info from the mds. Include leaf 242 * fragment in tree ONLY if ndist > 0. Otherwise, only 243 * branches/splits are included in i_fragtree) 244 */ 245 static int ceph_fill_dirfrag(struct inode *inode, 246 struct ceph_mds_reply_dirfrag *dirinfo) 247 { 248 struct ceph_inode_info *ci = ceph_inode(inode); 249 struct ceph_inode_frag *frag; 250 u32 id = le32_to_cpu(dirinfo->frag); 251 int mds = le32_to_cpu(dirinfo->auth); 252 int ndist = le32_to_cpu(dirinfo->ndist); 253 int diri_auth = -1; 254 int i; 255 int err = 0; 256 257 spin_lock(&ci->i_ceph_lock); 258 if (ci->i_auth_cap) 259 diri_auth = ci->i_auth_cap->mds; 260 spin_unlock(&ci->i_ceph_lock); 261 262 if (mds == -1) /* CDIR_AUTH_PARENT */ 263 mds = diri_auth; 264 265 mutex_lock(&ci->i_fragtree_mutex); 266 if (ndist == 0 && mds == diri_auth) { 267 /* no delegation info needed. */ 268 frag = __ceph_find_frag(ci, id); 269 if (!frag) 270 goto out; 271 if (frag->split_by == 0) { 272 /* tree leaf, remove */ 273 dout("fill_dirfrag removed %llx.%llx frag %x" 274 " (no ref)\n", ceph_vinop(inode), id); 275 rb_erase(&frag->node, &ci->i_fragtree); 276 kfree(frag); 277 } else { 278 /* tree branch, keep and clear */ 279 dout("fill_dirfrag cleared %llx.%llx frag %x" 280 " referral\n", ceph_vinop(inode), id); 281 frag->mds = -1; 282 frag->ndist = 0; 283 } 284 goto out; 285 } 286 287 288 /* find/add this frag to store mds delegation info */ 289 frag = __get_or_create_frag(ci, id); 290 if (IS_ERR(frag)) { 291 /* this is not the end of the world; we can continue 292 with bad/inaccurate delegation info */ 293 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n", 294 ceph_vinop(inode), le32_to_cpu(dirinfo->frag)); 295 err = -ENOMEM; 296 goto out; 297 } 298 299 frag->mds = mds; 300 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP); 301 for (i = 0; i < frag->ndist; i++) 302 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]); 303 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n", 304 ceph_vinop(inode), frag->frag, frag->ndist); 305 306 out: 307 mutex_unlock(&ci->i_fragtree_mutex); 308 return err; 309 } 310 311 static int frag_tree_split_cmp(const void *l, const void *r) 312 { 313 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l; 314 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r; 315 return ceph_frag_compare(le32_to_cpu(ls->frag), 316 le32_to_cpu(rs->frag)); 317 } 318 319 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag) 320 { 321 if (!frag) 322 return f == ceph_frag_make(0, 0); 323 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by) 324 return false; 325 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f)); 326 } 327 328 static int ceph_fill_fragtree(struct inode *inode, 329 struct ceph_frag_tree_head *fragtree, 330 struct ceph_mds_reply_dirfrag *dirinfo) 331 { 332 struct ceph_inode_info *ci = ceph_inode(inode); 333 struct ceph_inode_frag *frag, *prev_frag = NULL; 334 struct rb_node *rb_node; 335 unsigned i, split_by, nsplits; 336 u32 id; 337 bool update = false; 338 339 mutex_lock(&ci->i_fragtree_mutex); 340 nsplits = le32_to_cpu(fragtree->nsplits); 341 if (nsplits != ci->i_fragtree_nsplits) { 342 update = true; 343 } else if (nsplits) { 344 i = prandom_u32() % nsplits; 345 id = le32_to_cpu(fragtree->splits[i].frag); 346 if (!__ceph_find_frag(ci, id)) 347 update = true; 348 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) { 349 rb_node = rb_first(&ci->i_fragtree); 350 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 351 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node)) 352 update = true; 353 } 354 if (!update && dirinfo) { 355 id = le32_to_cpu(dirinfo->frag); 356 if (id != __ceph_choose_frag(ci, id, NULL, NULL)) 357 update = true; 358 } 359 if (!update) 360 goto out_unlock; 361 362 if (nsplits > 1) { 363 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]), 364 frag_tree_split_cmp, NULL); 365 } 366 367 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode)); 368 rb_node = rb_first(&ci->i_fragtree); 369 for (i = 0; i < nsplits; i++) { 370 id = le32_to_cpu(fragtree->splits[i].frag); 371 split_by = le32_to_cpu(fragtree->splits[i].by); 372 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) { 373 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, " 374 "frag %x split by %d\n", ceph_vinop(inode), 375 i, nsplits, id, split_by); 376 continue; 377 } 378 frag = NULL; 379 while (rb_node) { 380 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 381 if (ceph_frag_compare(frag->frag, id) >= 0) { 382 if (frag->frag != id) 383 frag = NULL; 384 else 385 rb_node = rb_next(rb_node); 386 break; 387 } 388 rb_node = rb_next(rb_node); 389 /* delete stale split/leaf node */ 390 if (frag->split_by > 0 || 391 !is_frag_child(frag->frag, prev_frag)) { 392 rb_erase(&frag->node, &ci->i_fragtree); 393 if (frag->split_by > 0) 394 ci->i_fragtree_nsplits--; 395 kfree(frag); 396 } 397 frag = NULL; 398 } 399 if (!frag) { 400 frag = __get_or_create_frag(ci, id); 401 if (IS_ERR(frag)) 402 continue; 403 } 404 if (frag->split_by == 0) 405 ci->i_fragtree_nsplits++; 406 frag->split_by = split_by; 407 dout(" frag %x split by %d\n", frag->frag, frag->split_by); 408 prev_frag = frag; 409 } 410 while (rb_node) { 411 frag = rb_entry(rb_node, struct ceph_inode_frag, node); 412 rb_node = rb_next(rb_node); 413 /* delete stale split/leaf node */ 414 if (frag->split_by > 0 || 415 !is_frag_child(frag->frag, prev_frag)) { 416 rb_erase(&frag->node, &ci->i_fragtree); 417 if (frag->split_by > 0) 418 ci->i_fragtree_nsplits--; 419 kfree(frag); 420 } 421 } 422 out_unlock: 423 mutex_unlock(&ci->i_fragtree_mutex); 424 return 0; 425 } 426 427 /* 428 * initialize a newly allocated inode. 429 */ 430 struct inode *ceph_alloc_inode(struct super_block *sb) 431 { 432 struct ceph_inode_info *ci; 433 int i; 434 435 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS); 436 if (!ci) 437 return NULL; 438 439 dout("alloc_inode %p\n", &ci->vfs_inode); 440 441 spin_lock_init(&ci->i_ceph_lock); 442 443 ci->i_version = 0; 444 ci->i_inline_version = 0; 445 ci->i_time_warp_seq = 0; 446 ci->i_ceph_flags = 0; 447 atomic64_set(&ci->i_ordered_count, 1); 448 atomic64_set(&ci->i_release_count, 1); 449 atomic64_set(&ci->i_complete_seq[0], 0); 450 atomic64_set(&ci->i_complete_seq[1], 0); 451 ci->i_symlink = NULL; 452 453 ci->i_max_bytes = 0; 454 ci->i_max_files = 0; 455 456 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout)); 457 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout)); 458 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL); 459 460 ci->i_fragtree = RB_ROOT; 461 mutex_init(&ci->i_fragtree_mutex); 462 463 ci->i_xattrs.blob = NULL; 464 ci->i_xattrs.prealloc_blob = NULL; 465 ci->i_xattrs.dirty = false; 466 ci->i_xattrs.index = RB_ROOT; 467 ci->i_xattrs.count = 0; 468 ci->i_xattrs.names_size = 0; 469 ci->i_xattrs.vals_size = 0; 470 ci->i_xattrs.version = 0; 471 ci->i_xattrs.index_version = 0; 472 473 ci->i_caps = RB_ROOT; 474 ci->i_auth_cap = NULL; 475 ci->i_dirty_caps = 0; 476 ci->i_flushing_caps = 0; 477 INIT_LIST_HEAD(&ci->i_dirty_item); 478 INIT_LIST_HEAD(&ci->i_flushing_item); 479 ci->i_prealloc_cap_flush = NULL; 480 INIT_LIST_HEAD(&ci->i_cap_flush_list); 481 init_waitqueue_head(&ci->i_cap_wq); 482 ci->i_hold_caps_max = 0; 483 INIT_LIST_HEAD(&ci->i_cap_delay_list); 484 INIT_LIST_HEAD(&ci->i_cap_snaps); 485 ci->i_head_snapc = NULL; 486 ci->i_snap_caps = 0; 487 488 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ; 489 for (i = 0; i < CEPH_FILE_MODE_BITS; i++) 490 ci->i_nr_by_mode[i] = 0; 491 492 mutex_init(&ci->i_truncate_mutex); 493 ci->i_truncate_seq = 0; 494 ci->i_truncate_size = 0; 495 ci->i_truncate_pending = 0; 496 497 ci->i_max_size = 0; 498 ci->i_reported_size = 0; 499 ci->i_wanted_max_size = 0; 500 ci->i_requested_max_size = 0; 501 502 ci->i_pin_ref = 0; 503 ci->i_rd_ref = 0; 504 ci->i_rdcache_ref = 0; 505 ci->i_wr_ref = 0; 506 ci->i_wb_ref = 0; 507 ci->i_fx_ref = 0; 508 ci->i_wrbuffer_ref = 0; 509 ci->i_wrbuffer_ref_head = 0; 510 atomic_set(&ci->i_filelock_ref, 0); 511 atomic_set(&ci->i_shared_gen, 1); 512 ci->i_rdcache_gen = 0; 513 ci->i_rdcache_revoking = 0; 514 515 INIT_LIST_HEAD(&ci->i_unsafe_dirops); 516 INIT_LIST_HEAD(&ci->i_unsafe_iops); 517 spin_lock_init(&ci->i_unsafe_lock); 518 519 ci->i_snap_realm = NULL; 520 INIT_LIST_HEAD(&ci->i_snap_realm_item); 521 INIT_LIST_HEAD(&ci->i_snap_flush_item); 522 523 INIT_WORK(&ci->i_work, ceph_inode_work); 524 ci->i_work_mask = 0; 525 memset(&ci->i_btime, '\0', sizeof(ci->i_btime)); 526 527 ceph_fscache_inode_init(ci); 528 529 ci->i_meta_err = 0; 530 531 return &ci->vfs_inode; 532 } 533 534 void ceph_free_inode(struct inode *inode) 535 { 536 struct ceph_inode_info *ci = ceph_inode(inode); 537 538 kfree(ci->i_symlink); 539 kmem_cache_free(ceph_inode_cachep, ci); 540 } 541 542 void ceph_evict_inode(struct inode *inode) 543 { 544 struct ceph_inode_info *ci = ceph_inode(inode); 545 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 546 struct ceph_inode_frag *frag; 547 struct rb_node *n; 548 549 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode)); 550 551 percpu_counter_dec(&mdsc->metric.total_inodes); 552 553 truncate_inode_pages_final(&inode->i_data); 554 clear_inode(inode); 555 556 ceph_fscache_unregister_inode_cookie(ci); 557 558 __ceph_remove_caps(ci); 559 560 if (__ceph_has_any_quota(ci)) 561 ceph_adjust_quota_realms_count(inode, false); 562 563 /* 564 * we may still have a snap_realm reference if there are stray 565 * caps in i_snap_caps. 566 */ 567 if (ci->i_snap_realm) { 568 if (ceph_snap(inode) == CEPH_NOSNAP) { 569 struct ceph_snap_realm *realm = ci->i_snap_realm; 570 dout(" dropping residual ref to snap realm %p\n", 571 realm); 572 spin_lock(&realm->inodes_with_caps_lock); 573 list_del_init(&ci->i_snap_realm_item); 574 ci->i_snap_realm = NULL; 575 if (realm->ino == ci->i_vino.ino) 576 realm->inode = NULL; 577 spin_unlock(&realm->inodes_with_caps_lock); 578 ceph_put_snap_realm(mdsc, realm); 579 } else { 580 ceph_put_snapid_map(mdsc, ci->i_snapid_map); 581 ci->i_snap_realm = NULL; 582 } 583 } 584 585 while ((n = rb_first(&ci->i_fragtree)) != NULL) { 586 frag = rb_entry(n, struct ceph_inode_frag, node); 587 rb_erase(n, &ci->i_fragtree); 588 kfree(frag); 589 } 590 ci->i_fragtree_nsplits = 0; 591 592 __ceph_destroy_xattrs(ci); 593 if (ci->i_xattrs.blob) 594 ceph_buffer_put(ci->i_xattrs.blob); 595 if (ci->i_xattrs.prealloc_blob) 596 ceph_buffer_put(ci->i_xattrs.prealloc_blob); 597 598 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns)); 599 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns)); 600 } 601 602 static inline blkcnt_t calc_inode_blocks(u64 size) 603 { 604 return (size + (1<<9) - 1) >> 9; 605 } 606 607 /* 608 * Helpers to fill in size, ctime, mtime, and atime. We have to be 609 * careful because either the client or MDS may have more up to date 610 * info, depending on which capabilities are held, and whether 611 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime 612 * and size are monotonically increasing, except when utimes() or 613 * truncate() increments the corresponding _seq values.) 614 */ 615 int ceph_fill_file_size(struct inode *inode, int issued, 616 u32 truncate_seq, u64 truncate_size, u64 size) 617 { 618 struct ceph_inode_info *ci = ceph_inode(inode); 619 int queue_trunc = 0; 620 loff_t isize = i_size_read(inode); 621 622 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 || 623 (truncate_seq == ci->i_truncate_seq && size > isize)) { 624 dout("size %lld -> %llu\n", isize, size); 625 if (size > 0 && S_ISDIR(inode->i_mode)) { 626 pr_err("fill_file_size non-zero size for directory\n"); 627 size = 0; 628 } 629 i_size_write(inode, size); 630 inode->i_blocks = calc_inode_blocks(size); 631 ci->i_reported_size = size; 632 if (truncate_seq != ci->i_truncate_seq) { 633 dout("truncate_seq %u -> %u\n", 634 ci->i_truncate_seq, truncate_seq); 635 ci->i_truncate_seq = truncate_seq; 636 637 /* the MDS should have revoked these caps */ 638 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL | 639 CEPH_CAP_FILE_RD | 640 CEPH_CAP_FILE_WR | 641 CEPH_CAP_FILE_LAZYIO)); 642 /* 643 * If we hold relevant caps, or in the case where we're 644 * not the only client referencing this file and we 645 * don't hold those caps, then we need to check whether 646 * the file is either opened or mmaped 647 */ 648 if ((issued & (CEPH_CAP_FILE_CACHE| 649 CEPH_CAP_FILE_BUFFER)) || 650 mapping_mapped(inode->i_mapping) || 651 __ceph_is_file_opened(ci)) { 652 ci->i_truncate_pending++; 653 queue_trunc = 1; 654 } 655 } 656 } 657 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 && 658 ci->i_truncate_size != truncate_size) { 659 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size, 660 truncate_size); 661 ci->i_truncate_size = truncate_size; 662 } 663 664 if (queue_trunc) 665 ceph_fscache_invalidate(inode); 666 667 return queue_trunc; 668 } 669 670 void ceph_fill_file_time(struct inode *inode, int issued, 671 u64 time_warp_seq, struct timespec64 *ctime, 672 struct timespec64 *mtime, struct timespec64 *atime) 673 { 674 struct ceph_inode_info *ci = ceph_inode(inode); 675 int warn = 0; 676 677 if (issued & (CEPH_CAP_FILE_EXCL| 678 CEPH_CAP_FILE_WR| 679 CEPH_CAP_FILE_BUFFER| 680 CEPH_CAP_AUTH_EXCL| 681 CEPH_CAP_XATTR_EXCL)) { 682 if (ci->i_version == 0 || 683 timespec64_compare(ctime, &inode->i_ctime) > 0) { 684 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n", 685 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 686 ctime->tv_sec, ctime->tv_nsec); 687 inode->i_ctime = *ctime; 688 } 689 if (ci->i_version == 0 || 690 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) { 691 /* the MDS did a utimes() */ 692 dout("mtime %lld.%09ld -> %lld.%09ld " 693 "tw %d -> %d\n", 694 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 695 mtime->tv_sec, mtime->tv_nsec, 696 ci->i_time_warp_seq, (int)time_warp_seq); 697 698 inode->i_mtime = *mtime; 699 inode->i_atime = *atime; 700 ci->i_time_warp_seq = time_warp_seq; 701 } else if (time_warp_seq == ci->i_time_warp_seq) { 702 /* nobody did utimes(); take the max */ 703 if (timespec64_compare(mtime, &inode->i_mtime) > 0) { 704 dout("mtime %lld.%09ld -> %lld.%09ld inc\n", 705 inode->i_mtime.tv_sec, 706 inode->i_mtime.tv_nsec, 707 mtime->tv_sec, mtime->tv_nsec); 708 inode->i_mtime = *mtime; 709 } 710 if (timespec64_compare(atime, &inode->i_atime) > 0) { 711 dout("atime %lld.%09ld -> %lld.%09ld inc\n", 712 inode->i_atime.tv_sec, 713 inode->i_atime.tv_nsec, 714 atime->tv_sec, atime->tv_nsec); 715 inode->i_atime = *atime; 716 } 717 } else if (issued & CEPH_CAP_FILE_EXCL) { 718 /* we did a utimes(); ignore mds values */ 719 } else { 720 warn = 1; 721 } 722 } else { 723 /* we have no write|excl caps; whatever the MDS says is true */ 724 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) { 725 inode->i_ctime = *ctime; 726 inode->i_mtime = *mtime; 727 inode->i_atime = *atime; 728 ci->i_time_warp_seq = time_warp_seq; 729 } else { 730 warn = 1; 731 } 732 } 733 if (warn) /* time_warp_seq shouldn't go backwards */ 734 dout("%p mds time_warp_seq %llu < %u\n", 735 inode, time_warp_seq, ci->i_time_warp_seq); 736 } 737 738 /* 739 * Populate an inode based on info from mds. May be called on new or 740 * existing inodes. 741 */ 742 int ceph_fill_inode(struct inode *inode, struct page *locked_page, 743 struct ceph_mds_reply_info_in *iinfo, 744 struct ceph_mds_reply_dirfrag *dirinfo, 745 struct ceph_mds_session *session, int cap_fmode, 746 struct ceph_cap_reservation *caps_reservation) 747 { 748 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb); 749 struct ceph_mds_reply_inode *info = iinfo->in; 750 struct ceph_inode_info *ci = ceph_inode(inode); 751 int issued, new_issued, info_caps; 752 struct timespec64 mtime, atime, ctime; 753 struct ceph_buffer *xattr_blob = NULL; 754 struct ceph_buffer *old_blob = NULL; 755 struct ceph_string *pool_ns = NULL; 756 struct ceph_cap *new_cap = NULL; 757 int err = 0; 758 bool wake = false; 759 bool queue_trunc = false; 760 bool new_version = false; 761 bool fill_inline = false; 762 763 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__, 764 inode, ceph_vinop(inode), le64_to_cpu(info->version), 765 ci->i_version); 766 767 info_caps = le32_to_cpu(info->cap.caps); 768 769 /* prealloc new cap struct */ 770 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) { 771 new_cap = ceph_get_cap(mdsc, caps_reservation); 772 if (!new_cap) 773 return -ENOMEM; 774 } 775 776 /* 777 * prealloc xattr data, if it looks like we'll need it. only 778 * if len > 4 (meaning there are actually xattrs; the first 4 779 * bytes are the xattr count). 780 */ 781 if (iinfo->xattr_len > 4) { 782 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS); 783 if (!xattr_blob) 784 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__, 785 iinfo->xattr_len); 786 } 787 788 if (iinfo->pool_ns_len > 0) 789 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data, 790 iinfo->pool_ns_len); 791 792 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map) 793 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode)); 794 795 spin_lock(&ci->i_ceph_lock); 796 797 /* 798 * provided version will be odd if inode value is projected, 799 * even if stable. skip the update if we have newer stable 800 * info (ours>=theirs, e.g. due to racing mds replies), unless 801 * we are getting projected (unstable) info (in which case the 802 * version is odd, and we want ours>theirs). 803 * us them 804 * 2 2 skip 805 * 3 2 skip 806 * 3 3 update 807 */ 808 if (ci->i_version == 0 || 809 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 810 le64_to_cpu(info->version) > (ci->i_version & ~1))) 811 new_version = true; 812 813 /* Update change_attribute */ 814 inode_set_max_iversion_raw(inode, iinfo->change_attr); 815 816 __ceph_caps_issued(ci, &issued); 817 issued |= __ceph_caps_dirty(ci); 818 new_issued = ~issued & info_caps; 819 820 /* update inode */ 821 inode->i_rdev = le32_to_cpu(info->rdev); 822 /* directories have fl_stripe_unit set to zero */ 823 if (le32_to_cpu(info->layout.fl_stripe_unit)) 824 inode->i_blkbits = 825 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1; 826 else 827 inode->i_blkbits = CEPH_BLOCK_SHIFT; 828 829 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files); 830 831 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && 832 (issued & CEPH_CAP_AUTH_EXCL) == 0) { 833 inode->i_mode = le32_to_cpu(info->mode); 834 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid)); 835 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid)); 836 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode, 837 from_kuid(&init_user_ns, inode->i_uid), 838 from_kgid(&init_user_ns, inode->i_gid)); 839 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime); 840 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime); 841 } 842 843 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && 844 (issued & CEPH_CAP_LINK_EXCL) == 0) 845 set_nlink(inode, le32_to_cpu(info->nlink)); 846 847 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { 848 /* be careful with mtime, atime, size */ 849 ceph_decode_timespec64(&atime, &info->atime); 850 ceph_decode_timespec64(&mtime, &info->mtime); 851 ceph_decode_timespec64(&ctime, &info->ctime); 852 ceph_fill_file_time(inode, issued, 853 le32_to_cpu(info->time_warp_seq), 854 &ctime, &mtime, &atime); 855 } 856 857 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) { 858 ci->i_files = le64_to_cpu(info->files); 859 ci->i_subdirs = le64_to_cpu(info->subdirs); 860 } 861 862 if (new_version || 863 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { 864 s64 old_pool = ci->i_layout.pool_id; 865 struct ceph_string *old_ns; 866 867 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout); 868 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns, 869 lockdep_is_held(&ci->i_ceph_lock)); 870 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns); 871 872 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns) 873 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM; 874 875 pool_ns = old_ns; 876 877 queue_trunc = ceph_fill_file_size(inode, issued, 878 le32_to_cpu(info->truncate_seq), 879 le64_to_cpu(info->truncate_size), 880 le64_to_cpu(info->size)); 881 /* only update max_size on auth cap */ 882 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) && 883 ci->i_max_size != le64_to_cpu(info->max_size)) { 884 dout("max_size %lld -> %llu\n", ci->i_max_size, 885 le64_to_cpu(info->max_size)); 886 ci->i_max_size = le64_to_cpu(info->max_size); 887 } 888 } 889 890 /* layout and rstat are not tracked by capability, update them if 891 * the inode info is from auth mds */ 892 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) { 893 if (S_ISDIR(inode->i_mode)) { 894 ci->i_dir_layout = iinfo->dir_layout; 895 ci->i_rbytes = le64_to_cpu(info->rbytes); 896 ci->i_rfiles = le64_to_cpu(info->rfiles); 897 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs); 898 ci->i_dir_pin = iinfo->dir_pin; 899 ci->i_rsnaps = iinfo->rsnaps; 900 ceph_decode_timespec64(&ci->i_rctime, &info->rctime); 901 } 902 } 903 904 /* xattrs */ 905 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */ 906 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && 907 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) { 908 if (ci->i_xattrs.blob) 909 old_blob = ci->i_xattrs.blob; 910 ci->i_xattrs.blob = xattr_blob; 911 if (xattr_blob) 912 memcpy(ci->i_xattrs.blob->vec.iov_base, 913 iinfo->xattr_data, iinfo->xattr_len); 914 ci->i_xattrs.version = le64_to_cpu(info->xattr_version); 915 ceph_forget_all_cached_acls(inode); 916 ceph_security_invalidate_secctx(inode); 917 xattr_blob = NULL; 918 } 919 920 /* finally update i_version */ 921 if (le64_to_cpu(info->version) > ci->i_version) 922 ci->i_version = le64_to_cpu(info->version); 923 924 inode->i_mapping->a_ops = &ceph_aops; 925 926 switch (inode->i_mode & S_IFMT) { 927 case S_IFIFO: 928 case S_IFBLK: 929 case S_IFCHR: 930 case S_IFSOCK: 931 inode->i_blkbits = PAGE_SHIFT; 932 init_special_inode(inode, inode->i_mode, inode->i_rdev); 933 inode->i_op = &ceph_file_iops; 934 break; 935 case S_IFREG: 936 inode->i_op = &ceph_file_iops; 937 inode->i_fop = &ceph_file_fops; 938 break; 939 case S_IFLNK: 940 inode->i_op = &ceph_symlink_iops; 941 if (!ci->i_symlink) { 942 u32 symlen = iinfo->symlink_len; 943 char *sym; 944 945 spin_unlock(&ci->i_ceph_lock); 946 947 if (symlen != i_size_read(inode)) { 948 pr_err("%s %llx.%llx BAD symlink " 949 "size %lld\n", __func__, 950 ceph_vinop(inode), 951 i_size_read(inode)); 952 i_size_write(inode, symlen); 953 inode->i_blocks = calc_inode_blocks(symlen); 954 } 955 956 err = -ENOMEM; 957 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS); 958 if (!sym) 959 goto out; 960 961 spin_lock(&ci->i_ceph_lock); 962 if (!ci->i_symlink) 963 ci->i_symlink = sym; 964 else 965 kfree(sym); /* lost a race */ 966 } 967 inode->i_link = ci->i_symlink; 968 break; 969 case S_IFDIR: 970 inode->i_op = &ceph_dir_iops; 971 inode->i_fop = &ceph_dir_fops; 972 break; 973 default: 974 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__, 975 ceph_vinop(inode), inode->i_mode); 976 } 977 978 /* were we issued a capability? */ 979 if (info_caps) { 980 if (ceph_snap(inode) == CEPH_NOSNAP) { 981 ceph_add_cap(inode, session, 982 le64_to_cpu(info->cap.cap_id), 983 info_caps, 984 le32_to_cpu(info->cap.wanted), 985 le32_to_cpu(info->cap.seq), 986 le32_to_cpu(info->cap.mseq), 987 le64_to_cpu(info->cap.realm), 988 info->cap.flags, &new_cap); 989 990 /* set dir completion flag? */ 991 if (S_ISDIR(inode->i_mode) && 992 ci->i_files == 0 && ci->i_subdirs == 0 && 993 (info_caps & CEPH_CAP_FILE_SHARED) && 994 (issued & CEPH_CAP_FILE_EXCL) == 0 && 995 !__ceph_dir_is_complete(ci)) { 996 dout(" marking %p complete (empty)\n", inode); 997 i_size_write(inode, 0); 998 __ceph_dir_set_complete(ci, 999 atomic64_read(&ci->i_release_count), 1000 atomic64_read(&ci->i_ordered_count)); 1001 } 1002 1003 wake = true; 1004 } else { 1005 dout(" %p got snap_caps %s\n", inode, 1006 ceph_cap_string(info_caps)); 1007 ci->i_snap_caps |= info_caps; 1008 } 1009 } 1010 1011 if (iinfo->inline_version > 0 && 1012 iinfo->inline_version >= ci->i_inline_version) { 1013 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; 1014 ci->i_inline_version = iinfo->inline_version; 1015 if (ci->i_inline_version != CEPH_INLINE_NONE && 1016 (locked_page || (info_caps & cache_caps))) 1017 fill_inline = true; 1018 } 1019 1020 if (cap_fmode >= 0) { 1021 if (!info_caps) 1022 pr_warn("mds issued no caps on %llx.%llx\n", 1023 ceph_vinop(inode)); 1024 __ceph_touch_fmode(ci, mdsc, cap_fmode); 1025 } 1026 1027 spin_unlock(&ci->i_ceph_lock); 1028 1029 if (fill_inline) 1030 ceph_fill_inline_data(inode, locked_page, 1031 iinfo->inline_data, iinfo->inline_len); 1032 1033 if (wake) 1034 wake_up_all(&ci->i_cap_wq); 1035 1036 /* queue truncate if we saw i_size decrease */ 1037 if (queue_trunc) 1038 ceph_queue_vmtruncate(inode); 1039 1040 /* populate frag tree */ 1041 if (S_ISDIR(inode->i_mode)) 1042 ceph_fill_fragtree(inode, &info->fragtree, dirinfo); 1043 1044 /* update delegation info? */ 1045 if (dirinfo) 1046 ceph_fill_dirfrag(inode, dirinfo); 1047 1048 err = 0; 1049 out: 1050 if (new_cap) 1051 ceph_put_cap(mdsc, new_cap); 1052 ceph_buffer_put(old_blob); 1053 ceph_buffer_put(xattr_blob); 1054 ceph_put_string(pool_ns); 1055 return err; 1056 } 1057 1058 /* 1059 * caller should hold session s_mutex and dentry->d_lock. 1060 */ 1061 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry, 1062 struct ceph_mds_reply_lease *lease, 1063 struct ceph_mds_session *session, 1064 unsigned long from_time, 1065 struct ceph_mds_session **old_lease_session) 1066 { 1067 struct ceph_dentry_info *di = ceph_dentry(dentry); 1068 unsigned mask = le16_to_cpu(lease->mask); 1069 long unsigned duration = le32_to_cpu(lease->duration_ms); 1070 long unsigned ttl = from_time + (duration * HZ) / 1000; 1071 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000; 1072 1073 dout("update_dentry_lease %p duration %lu ms ttl %lu\n", 1074 dentry, duration, ttl); 1075 1076 /* only track leases on regular dentries */ 1077 if (ceph_snap(dir) != CEPH_NOSNAP) 1078 return; 1079 1080 if (mask & CEPH_LEASE_PRIMARY_LINK) 1081 di->flags |= CEPH_DENTRY_PRIMARY_LINK; 1082 else 1083 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK; 1084 1085 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen); 1086 if (!(mask & CEPH_LEASE_VALID)) { 1087 __ceph_dentry_dir_lease_touch(di); 1088 return; 1089 } 1090 1091 if (di->lease_gen == session->s_cap_gen && 1092 time_before(ttl, di->time)) 1093 return; /* we already have a newer lease. */ 1094 1095 if (di->lease_session && di->lease_session != session) { 1096 *old_lease_session = di->lease_session; 1097 di->lease_session = NULL; 1098 } 1099 1100 if (!di->lease_session) 1101 di->lease_session = ceph_get_mds_session(session); 1102 di->lease_gen = session->s_cap_gen; 1103 di->lease_seq = le32_to_cpu(lease->seq); 1104 di->lease_renew_after = half_ttl; 1105 di->lease_renew_from = 0; 1106 di->time = ttl; 1107 1108 __ceph_dentry_lease_touch(di); 1109 } 1110 1111 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry, 1112 struct ceph_mds_reply_lease *lease, 1113 struct ceph_mds_session *session, 1114 unsigned long from_time) 1115 { 1116 struct ceph_mds_session *old_lease_session = NULL; 1117 spin_lock(&dentry->d_lock); 1118 __update_dentry_lease(dir, dentry, lease, session, from_time, 1119 &old_lease_session); 1120 spin_unlock(&dentry->d_lock); 1121 if (old_lease_session) 1122 ceph_put_mds_session(old_lease_session); 1123 } 1124 1125 /* 1126 * update dentry lease without having parent inode locked 1127 */ 1128 static void update_dentry_lease_careful(struct dentry *dentry, 1129 struct ceph_mds_reply_lease *lease, 1130 struct ceph_mds_session *session, 1131 unsigned long from_time, 1132 char *dname, u32 dname_len, 1133 struct ceph_vino *pdvino, 1134 struct ceph_vino *ptvino) 1135 1136 { 1137 struct inode *dir; 1138 struct ceph_mds_session *old_lease_session = NULL; 1139 1140 spin_lock(&dentry->d_lock); 1141 /* make sure dentry's name matches target */ 1142 if (dentry->d_name.len != dname_len || 1143 memcmp(dentry->d_name.name, dname, dname_len)) 1144 goto out_unlock; 1145 1146 dir = d_inode(dentry->d_parent); 1147 /* make sure parent matches dvino */ 1148 if (!ceph_ino_compare(dir, pdvino)) 1149 goto out_unlock; 1150 1151 /* make sure dentry's inode matches target. NULL ptvino means that 1152 * we expect a negative dentry */ 1153 if (ptvino) { 1154 if (d_really_is_negative(dentry)) 1155 goto out_unlock; 1156 if (!ceph_ino_compare(d_inode(dentry), ptvino)) 1157 goto out_unlock; 1158 } else { 1159 if (d_really_is_positive(dentry)) 1160 goto out_unlock; 1161 } 1162 1163 __update_dentry_lease(dir, dentry, lease, session, 1164 from_time, &old_lease_session); 1165 out_unlock: 1166 spin_unlock(&dentry->d_lock); 1167 if (old_lease_session) 1168 ceph_put_mds_session(old_lease_session); 1169 } 1170 1171 /* 1172 * splice a dentry to an inode. 1173 * caller must hold directory i_mutex for this to be safe. 1174 */ 1175 static int splice_dentry(struct dentry **pdn, struct inode *in) 1176 { 1177 struct dentry *dn = *pdn; 1178 struct dentry *realdn; 1179 1180 BUG_ON(d_inode(dn)); 1181 1182 if (S_ISDIR(in->i_mode)) { 1183 /* If inode is directory, d_splice_alias() below will remove 1184 * 'realdn' from its origin parent. We need to ensure that 1185 * origin parent's readdir cache will not reference 'realdn' 1186 */ 1187 realdn = d_find_any_alias(in); 1188 if (realdn) { 1189 struct ceph_dentry_info *di = ceph_dentry(realdn); 1190 spin_lock(&realdn->d_lock); 1191 1192 realdn->d_op->d_prune(realdn); 1193 1194 di->time = jiffies; 1195 di->lease_shared_gen = 0; 1196 di->offset = 0; 1197 1198 spin_unlock(&realdn->d_lock); 1199 dput(realdn); 1200 } 1201 } 1202 1203 /* dn must be unhashed */ 1204 if (!d_unhashed(dn)) 1205 d_drop(dn); 1206 realdn = d_splice_alias(in, dn); 1207 if (IS_ERR(realdn)) { 1208 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n", 1209 PTR_ERR(realdn), dn, in, ceph_vinop(in)); 1210 return PTR_ERR(realdn); 1211 } 1212 1213 if (realdn) { 1214 dout("dn %p (%d) spliced with %p (%d) " 1215 "inode %p ino %llx.%llx\n", 1216 dn, d_count(dn), 1217 realdn, d_count(realdn), 1218 d_inode(realdn), ceph_vinop(d_inode(realdn))); 1219 dput(dn); 1220 *pdn = realdn; 1221 } else { 1222 BUG_ON(!ceph_dentry(dn)); 1223 dout("dn %p attached to %p ino %llx.%llx\n", 1224 dn, d_inode(dn), ceph_vinop(d_inode(dn))); 1225 } 1226 return 0; 1227 } 1228 1229 /* 1230 * Incorporate results into the local cache. This is either just 1231 * one inode, or a directory, dentry, and possibly linked-to inode (e.g., 1232 * after a lookup). 1233 * 1234 * A reply may contain 1235 * a directory inode along with a dentry. 1236 * and/or a target inode 1237 * 1238 * Called with snap_rwsem (read). 1239 */ 1240 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req) 1241 { 1242 struct ceph_mds_session *session = req->r_session; 1243 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1244 struct inode *in = NULL; 1245 struct ceph_vino tvino, dvino; 1246 struct ceph_fs_client *fsc = ceph_sb_to_client(sb); 1247 int err = 0; 1248 1249 dout("fill_trace %p is_dentry %d is_target %d\n", req, 1250 rinfo->head->is_dentry, rinfo->head->is_target); 1251 1252 if (!rinfo->head->is_target && !rinfo->head->is_dentry) { 1253 dout("fill_trace reply is empty!\n"); 1254 if (rinfo->head->result == 0 && req->r_parent) 1255 ceph_invalidate_dir_request(req); 1256 return 0; 1257 } 1258 1259 if (rinfo->head->is_dentry) { 1260 struct inode *dir = req->r_parent; 1261 1262 if (dir) { 1263 err = ceph_fill_inode(dir, NULL, &rinfo->diri, 1264 rinfo->dirfrag, session, -1, 1265 &req->r_caps_reservation); 1266 if (err < 0) 1267 goto done; 1268 } else { 1269 WARN_ON_ONCE(1); 1270 } 1271 1272 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME && 1273 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1274 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1275 struct qstr dname; 1276 struct dentry *dn, *parent; 1277 1278 BUG_ON(!rinfo->head->is_target); 1279 BUG_ON(req->r_dentry); 1280 1281 parent = d_find_any_alias(dir); 1282 BUG_ON(!parent); 1283 1284 dname.name = rinfo->dname; 1285 dname.len = rinfo->dname_len; 1286 dname.hash = full_name_hash(parent, dname.name, dname.len); 1287 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1288 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1289 retry_lookup: 1290 dn = d_lookup(parent, &dname); 1291 dout("d_lookup on parent=%p name=%.*s got %p\n", 1292 parent, dname.len, dname.name, dn); 1293 1294 if (!dn) { 1295 dn = d_alloc(parent, &dname); 1296 dout("d_alloc %p '%.*s' = %p\n", parent, 1297 dname.len, dname.name, dn); 1298 if (!dn) { 1299 dput(parent); 1300 err = -ENOMEM; 1301 goto done; 1302 } 1303 err = 0; 1304 } else if (d_really_is_positive(dn) && 1305 (ceph_ino(d_inode(dn)) != tvino.ino || 1306 ceph_snap(d_inode(dn)) != tvino.snap)) { 1307 dout(" dn %p points to wrong inode %p\n", 1308 dn, d_inode(dn)); 1309 ceph_dir_clear_ordered(dir); 1310 d_delete(dn); 1311 dput(dn); 1312 goto retry_lookup; 1313 } 1314 1315 req->r_dentry = dn; 1316 dput(parent); 1317 } 1318 } 1319 1320 if (rinfo->head->is_target) { 1321 /* Should be filled in by handle_reply */ 1322 BUG_ON(!req->r_target_inode); 1323 1324 in = req->r_target_inode; 1325 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti, 1326 NULL, session, 1327 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1328 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) && 1329 rinfo->head->result == 0) ? req->r_fmode : -1, 1330 &req->r_caps_reservation); 1331 if (err < 0) { 1332 pr_err("ceph_fill_inode badness %p %llx.%llx\n", 1333 in, ceph_vinop(in)); 1334 req->r_target_inode = NULL; 1335 if (in->i_state & I_NEW) 1336 discard_new_inode(in); 1337 else 1338 iput(in); 1339 goto done; 1340 } 1341 if (in->i_state & I_NEW) 1342 unlock_new_inode(in); 1343 } 1344 1345 /* 1346 * ignore null lease/binding on snapdir ENOENT, or else we 1347 * will have trouble splicing in the virtual snapdir later 1348 */ 1349 if (rinfo->head->is_dentry && 1350 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) && 1351 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1352 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name, 1353 fsc->mount_options->snapdir_name, 1354 req->r_dentry->d_name.len))) { 1355 /* 1356 * lookup link rename : null -> possibly existing inode 1357 * mknod symlink mkdir : null -> new inode 1358 * unlink : linked -> null 1359 */ 1360 struct inode *dir = req->r_parent; 1361 struct dentry *dn = req->r_dentry; 1362 bool have_dir_cap, have_lease; 1363 1364 BUG_ON(!dn); 1365 BUG_ON(!dir); 1366 BUG_ON(d_inode(dn->d_parent) != dir); 1367 1368 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1369 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1370 1371 BUG_ON(ceph_ino(dir) != dvino.ino); 1372 BUG_ON(ceph_snap(dir) != dvino.snap); 1373 1374 /* do we have a lease on the whole dir? */ 1375 have_dir_cap = 1376 (le32_to_cpu(rinfo->diri.in->cap.caps) & 1377 CEPH_CAP_FILE_SHARED); 1378 1379 /* do we have a dn lease? */ 1380 have_lease = have_dir_cap || 1381 le32_to_cpu(rinfo->dlease->duration_ms); 1382 if (!have_lease) 1383 dout("fill_trace no dentry lease or dir cap\n"); 1384 1385 /* rename? */ 1386 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) { 1387 struct inode *olddir = req->r_old_dentry_dir; 1388 BUG_ON(!olddir); 1389 1390 dout(" src %p '%pd' dst %p '%pd'\n", 1391 req->r_old_dentry, 1392 req->r_old_dentry, 1393 dn, dn); 1394 dout("fill_trace doing d_move %p -> %p\n", 1395 req->r_old_dentry, dn); 1396 1397 /* d_move screws up sibling dentries' offsets */ 1398 ceph_dir_clear_ordered(dir); 1399 ceph_dir_clear_ordered(olddir); 1400 1401 d_move(req->r_old_dentry, dn); 1402 dout(" src %p '%pd' dst %p '%pd'\n", 1403 req->r_old_dentry, 1404 req->r_old_dentry, 1405 dn, dn); 1406 1407 /* ensure target dentry is invalidated, despite 1408 rehashing bug in vfs_rename_dir */ 1409 ceph_invalidate_dentry_lease(dn); 1410 1411 dout("dn %p gets new offset %lld\n", req->r_old_dentry, 1412 ceph_dentry(req->r_old_dentry)->offset); 1413 1414 /* swap r_dentry and r_old_dentry in case that 1415 * splice_dentry() gets called later. This is safe 1416 * because no other place will use them */ 1417 req->r_dentry = req->r_old_dentry; 1418 req->r_old_dentry = dn; 1419 dn = req->r_dentry; 1420 } 1421 1422 /* null dentry? */ 1423 if (!rinfo->head->is_target) { 1424 dout("fill_trace null dentry\n"); 1425 if (d_really_is_positive(dn)) { 1426 dout("d_delete %p\n", dn); 1427 ceph_dir_clear_ordered(dir); 1428 d_delete(dn); 1429 } else if (have_lease) { 1430 if (d_unhashed(dn)) 1431 d_add(dn, NULL); 1432 update_dentry_lease(dir, dn, 1433 rinfo->dlease, session, 1434 req->r_request_started); 1435 } 1436 goto done; 1437 } 1438 1439 /* attach proper inode */ 1440 if (d_really_is_negative(dn)) { 1441 ceph_dir_clear_ordered(dir); 1442 ihold(in); 1443 err = splice_dentry(&req->r_dentry, in); 1444 if (err < 0) 1445 goto done; 1446 dn = req->r_dentry; /* may have spliced */ 1447 } else if (d_really_is_positive(dn) && d_inode(dn) != in) { 1448 dout(" %p links to %p %llx.%llx, not %llx.%llx\n", 1449 dn, d_inode(dn), ceph_vinop(d_inode(dn)), 1450 ceph_vinop(in)); 1451 d_invalidate(dn); 1452 have_lease = false; 1453 } 1454 1455 if (have_lease) { 1456 update_dentry_lease(dir, dn, 1457 rinfo->dlease, session, 1458 req->r_request_started); 1459 } 1460 dout(" final dn %p\n", dn); 1461 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP || 1462 req->r_op == CEPH_MDS_OP_MKSNAP) && 1463 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) && 1464 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) { 1465 struct inode *dir = req->r_parent; 1466 1467 /* fill out a snapdir LOOKUPSNAP dentry */ 1468 BUG_ON(!dir); 1469 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR); 1470 BUG_ON(!req->r_dentry); 1471 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry); 1472 ceph_dir_clear_ordered(dir); 1473 ihold(in); 1474 err = splice_dentry(&req->r_dentry, in); 1475 if (err < 0) 1476 goto done; 1477 } else if (rinfo->head->is_dentry && req->r_dentry) { 1478 /* parent inode is not locked, be carefull */ 1479 struct ceph_vino *ptvino = NULL; 1480 dvino.ino = le64_to_cpu(rinfo->diri.in->ino); 1481 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid); 1482 if (rinfo->head->is_target) { 1483 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino); 1484 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid); 1485 ptvino = &tvino; 1486 } 1487 update_dentry_lease_careful(req->r_dentry, rinfo->dlease, 1488 session, req->r_request_started, 1489 rinfo->dname, rinfo->dname_len, 1490 &dvino, ptvino); 1491 } 1492 done: 1493 dout("fill_trace done err=%d\n", err); 1494 return err; 1495 } 1496 1497 /* 1498 * Prepopulate our cache with readdir results, leases, etc. 1499 */ 1500 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req, 1501 struct ceph_mds_session *session) 1502 { 1503 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1504 int i, err = 0; 1505 1506 for (i = 0; i < rinfo->dir_nr; i++) { 1507 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1508 struct ceph_vino vino; 1509 struct inode *in; 1510 int rc; 1511 1512 vino.ino = le64_to_cpu(rde->inode.in->ino); 1513 vino.snap = le64_to_cpu(rde->inode.in->snapid); 1514 1515 in = ceph_get_inode(req->r_dentry->d_sb, vino); 1516 if (IS_ERR(in)) { 1517 err = PTR_ERR(in); 1518 dout("new_inode badness got %d\n", err); 1519 continue; 1520 } 1521 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1522 -1, &req->r_caps_reservation); 1523 if (rc < 0) { 1524 pr_err("ceph_fill_inode badness on %p got %d\n", 1525 in, rc); 1526 err = rc; 1527 if (in->i_state & I_NEW) { 1528 ihold(in); 1529 discard_new_inode(in); 1530 } 1531 } else if (in->i_state & I_NEW) { 1532 unlock_new_inode(in); 1533 } 1534 1535 /* avoid calling iput_final() in mds dispatch threads */ 1536 ceph_async_iput(in); 1537 } 1538 1539 return err; 1540 } 1541 1542 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl) 1543 { 1544 if (ctl->page) { 1545 kunmap(ctl->page); 1546 put_page(ctl->page); 1547 ctl->page = NULL; 1548 } 1549 } 1550 1551 static int fill_readdir_cache(struct inode *dir, struct dentry *dn, 1552 struct ceph_readdir_cache_control *ctl, 1553 struct ceph_mds_request *req) 1554 { 1555 struct ceph_inode_info *ci = ceph_inode(dir); 1556 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*); 1557 unsigned idx = ctl->index % nsize; 1558 pgoff_t pgoff = ctl->index / nsize; 1559 1560 if (!ctl->page || pgoff != page_index(ctl->page)) { 1561 ceph_readdir_cache_release(ctl); 1562 if (idx == 0) 1563 ctl->page = grab_cache_page(&dir->i_data, pgoff); 1564 else 1565 ctl->page = find_lock_page(&dir->i_data, pgoff); 1566 if (!ctl->page) { 1567 ctl->index = -1; 1568 return idx == 0 ? -ENOMEM : 0; 1569 } 1570 /* reading/filling the cache are serialized by 1571 * i_mutex, no need to use page lock */ 1572 unlock_page(ctl->page); 1573 ctl->dentries = kmap(ctl->page); 1574 if (idx == 0) 1575 memset(ctl->dentries, 0, PAGE_SIZE); 1576 } 1577 1578 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) && 1579 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) { 1580 dout("readdir cache dn %p idx %d\n", dn, ctl->index); 1581 ctl->dentries[idx] = dn; 1582 ctl->index++; 1583 } else { 1584 dout("disable readdir cache\n"); 1585 ctl->index = -1; 1586 } 1587 return 0; 1588 } 1589 1590 int ceph_readdir_prepopulate(struct ceph_mds_request *req, 1591 struct ceph_mds_session *session) 1592 { 1593 struct dentry *parent = req->r_dentry; 1594 struct ceph_inode_info *ci = ceph_inode(d_inode(parent)); 1595 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info; 1596 struct qstr dname; 1597 struct dentry *dn; 1598 struct inode *in; 1599 int err = 0, skipped = 0, ret, i; 1600 u32 frag = le32_to_cpu(req->r_args.readdir.frag); 1601 u32 last_hash = 0; 1602 u32 fpos_offset; 1603 struct ceph_readdir_cache_control cache_ctl = {}; 1604 1605 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) 1606 return readdir_prepopulate_inodes_only(req, session); 1607 1608 if (rinfo->hash_order) { 1609 if (req->r_path2) { 1610 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1611 req->r_path2, 1612 strlen(req->r_path2)); 1613 last_hash = ceph_frag_value(last_hash); 1614 } else if (rinfo->offset_hash) { 1615 /* mds understands offset_hash */ 1616 WARN_ON_ONCE(req->r_readdir_offset != 2); 1617 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash); 1618 } 1619 } 1620 1621 if (rinfo->dir_dir && 1622 le32_to_cpu(rinfo->dir_dir->frag) != frag) { 1623 dout("readdir_prepopulate got new frag %x -> %x\n", 1624 frag, le32_to_cpu(rinfo->dir_dir->frag)); 1625 frag = le32_to_cpu(rinfo->dir_dir->frag); 1626 if (!rinfo->hash_order) 1627 req->r_readdir_offset = 2; 1628 } 1629 1630 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) { 1631 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n", 1632 rinfo->dir_nr, parent); 1633 } else { 1634 dout("readdir_prepopulate %d items under dn %p\n", 1635 rinfo->dir_nr, parent); 1636 if (rinfo->dir_dir) 1637 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir); 1638 1639 if (ceph_frag_is_leftmost(frag) && 1640 req->r_readdir_offset == 2 && 1641 !(rinfo->hash_order && last_hash)) { 1642 /* note dir version at start of readdir so we can 1643 * tell if any dentries get dropped */ 1644 req->r_dir_release_cnt = 1645 atomic64_read(&ci->i_release_count); 1646 req->r_dir_ordered_cnt = 1647 atomic64_read(&ci->i_ordered_count); 1648 req->r_readdir_cache_idx = 0; 1649 } 1650 } 1651 1652 cache_ctl.index = req->r_readdir_cache_idx; 1653 fpos_offset = req->r_readdir_offset; 1654 1655 /* FIXME: release caps/leases if error occurs */ 1656 for (i = 0; i < rinfo->dir_nr; i++) { 1657 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i; 1658 struct ceph_vino tvino; 1659 1660 dname.name = rde->name; 1661 dname.len = rde->name_len; 1662 dname.hash = full_name_hash(parent, dname.name, dname.len); 1663 1664 tvino.ino = le64_to_cpu(rde->inode.in->ino); 1665 tvino.snap = le64_to_cpu(rde->inode.in->snapid); 1666 1667 if (rinfo->hash_order) { 1668 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash, 1669 rde->name, rde->name_len); 1670 hash = ceph_frag_value(hash); 1671 if (hash != last_hash) 1672 fpos_offset = 2; 1673 last_hash = hash; 1674 rde->offset = ceph_make_fpos(hash, fpos_offset++, true); 1675 } else { 1676 rde->offset = ceph_make_fpos(frag, fpos_offset++, false); 1677 } 1678 1679 retry_lookup: 1680 dn = d_lookup(parent, &dname); 1681 dout("d_lookup on parent=%p name=%.*s got %p\n", 1682 parent, dname.len, dname.name, dn); 1683 1684 if (!dn) { 1685 dn = d_alloc(parent, &dname); 1686 dout("d_alloc %p '%.*s' = %p\n", parent, 1687 dname.len, dname.name, dn); 1688 if (!dn) { 1689 dout("d_alloc badness\n"); 1690 err = -ENOMEM; 1691 goto out; 1692 } 1693 } else if (d_really_is_positive(dn) && 1694 (ceph_ino(d_inode(dn)) != tvino.ino || 1695 ceph_snap(d_inode(dn)) != tvino.snap)) { 1696 struct ceph_dentry_info *di = ceph_dentry(dn); 1697 dout(" dn %p points to wrong inode %p\n", 1698 dn, d_inode(dn)); 1699 1700 spin_lock(&dn->d_lock); 1701 if (di->offset > 0 && 1702 di->lease_shared_gen == 1703 atomic_read(&ci->i_shared_gen)) { 1704 __ceph_dir_clear_ordered(ci); 1705 di->offset = 0; 1706 } 1707 spin_unlock(&dn->d_lock); 1708 1709 d_delete(dn); 1710 dput(dn); 1711 goto retry_lookup; 1712 } 1713 1714 /* inode */ 1715 if (d_really_is_positive(dn)) { 1716 in = d_inode(dn); 1717 } else { 1718 in = ceph_get_inode(parent->d_sb, tvino); 1719 if (IS_ERR(in)) { 1720 dout("new_inode badness\n"); 1721 d_drop(dn); 1722 dput(dn); 1723 err = PTR_ERR(in); 1724 goto out; 1725 } 1726 } 1727 1728 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session, 1729 -1, &req->r_caps_reservation); 1730 if (ret < 0) { 1731 pr_err("ceph_fill_inode badness on %p\n", in); 1732 if (d_really_is_negative(dn)) { 1733 /* avoid calling iput_final() in mds 1734 * dispatch threads */ 1735 if (in->i_state & I_NEW) { 1736 ihold(in); 1737 discard_new_inode(in); 1738 } 1739 ceph_async_iput(in); 1740 } 1741 d_drop(dn); 1742 err = ret; 1743 goto next_item; 1744 } 1745 if (in->i_state & I_NEW) 1746 unlock_new_inode(in); 1747 1748 if (d_really_is_negative(dn)) { 1749 if (ceph_security_xattr_deadlock(in)) { 1750 dout(" skip splicing dn %p to inode %p" 1751 " (security xattr deadlock)\n", dn, in); 1752 ceph_async_iput(in); 1753 skipped++; 1754 goto next_item; 1755 } 1756 1757 err = splice_dentry(&dn, in); 1758 if (err < 0) 1759 goto next_item; 1760 } 1761 1762 ceph_dentry(dn)->offset = rde->offset; 1763 1764 update_dentry_lease(d_inode(parent), dn, 1765 rde->lease, req->r_session, 1766 req->r_request_started); 1767 1768 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) { 1769 ret = fill_readdir_cache(d_inode(parent), dn, 1770 &cache_ctl, req); 1771 if (ret < 0) 1772 err = ret; 1773 } 1774 next_item: 1775 dput(dn); 1776 } 1777 out: 1778 if (err == 0 && skipped == 0) { 1779 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags); 1780 req->r_readdir_cache_idx = cache_ctl.index; 1781 } 1782 ceph_readdir_cache_release(&cache_ctl); 1783 dout("readdir_prepopulate done\n"); 1784 return err; 1785 } 1786 1787 bool ceph_inode_set_size(struct inode *inode, loff_t size) 1788 { 1789 struct ceph_inode_info *ci = ceph_inode(inode); 1790 bool ret; 1791 1792 spin_lock(&ci->i_ceph_lock); 1793 dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size); 1794 i_size_write(inode, size); 1795 inode->i_blocks = calc_inode_blocks(size); 1796 1797 ret = __ceph_should_report_size(ci); 1798 1799 spin_unlock(&ci->i_ceph_lock); 1800 return ret; 1801 } 1802 1803 /* 1804 * Put reference to inode, but avoid calling iput_final() in current thread. 1805 * iput_final() may wait for reahahead pages. The wait can cause deadlock in 1806 * some contexts. 1807 */ 1808 void ceph_async_iput(struct inode *inode) 1809 { 1810 if (!inode) 1811 return; 1812 for (;;) { 1813 if (atomic_add_unless(&inode->i_count, -1, 1)) 1814 break; 1815 if (queue_work(ceph_inode_to_client(inode)->inode_wq, 1816 &ceph_inode(inode)->i_work)) 1817 break; 1818 /* queue work failed, i_count must be at least 2 */ 1819 } 1820 } 1821 1822 void ceph_queue_inode_work(struct inode *inode, int work_bit) 1823 { 1824 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1825 struct ceph_inode_info *ci = ceph_inode(inode); 1826 set_bit(work_bit, &ci->i_work_mask); 1827 1828 ihold(inode); 1829 if (queue_work(fsc->inode_wq, &ci->i_work)) { 1830 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask); 1831 } else { 1832 dout("queue_inode_work %p already queued, mask=%lx\n", 1833 inode, ci->i_work_mask); 1834 iput(inode); 1835 } 1836 } 1837 1838 static void ceph_do_invalidate_pages(struct inode *inode) 1839 { 1840 struct ceph_inode_info *ci = ceph_inode(inode); 1841 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 1842 u32 orig_gen; 1843 int check = 0; 1844 1845 mutex_lock(&ci->i_truncate_mutex); 1846 1847 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) { 1848 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n", 1849 inode, ceph_ino(inode)); 1850 mapping_set_error(inode->i_mapping, -EIO); 1851 truncate_pagecache(inode, 0); 1852 mutex_unlock(&ci->i_truncate_mutex); 1853 goto out; 1854 } 1855 1856 spin_lock(&ci->i_ceph_lock); 1857 dout("invalidate_pages %p gen %d revoking %d\n", inode, 1858 ci->i_rdcache_gen, ci->i_rdcache_revoking); 1859 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) { 1860 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1861 check = 1; 1862 spin_unlock(&ci->i_ceph_lock); 1863 mutex_unlock(&ci->i_truncate_mutex); 1864 goto out; 1865 } 1866 orig_gen = ci->i_rdcache_gen; 1867 spin_unlock(&ci->i_ceph_lock); 1868 1869 ceph_fscache_invalidate(inode); 1870 if (invalidate_inode_pages2(inode->i_mapping) < 0) { 1871 pr_err("invalidate_pages %p fails\n", inode); 1872 } 1873 1874 spin_lock(&ci->i_ceph_lock); 1875 if (orig_gen == ci->i_rdcache_gen && 1876 orig_gen == ci->i_rdcache_revoking) { 1877 dout("invalidate_pages %p gen %d successful\n", inode, 1878 ci->i_rdcache_gen); 1879 ci->i_rdcache_revoking--; 1880 check = 1; 1881 } else { 1882 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n", 1883 inode, orig_gen, ci->i_rdcache_gen, 1884 ci->i_rdcache_revoking); 1885 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE)) 1886 check = 1; 1887 } 1888 spin_unlock(&ci->i_ceph_lock); 1889 mutex_unlock(&ci->i_truncate_mutex); 1890 out: 1891 if (check) 1892 ceph_check_caps(ci, 0, NULL); 1893 } 1894 1895 /* 1896 * Make sure any pending truncation is applied before doing anything 1897 * that may depend on it. 1898 */ 1899 void __ceph_do_pending_vmtruncate(struct inode *inode) 1900 { 1901 struct ceph_inode_info *ci = ceph_inode(inode); 1902 u64 to; 1903 int wrbuffer_refs, finish = 0; 1904 1905 mutex_lock(&ci->i_truncate_mutex); 1906 retry: 1907 spin_lock(&ci->i_ceph_lock); 1908 if (ci->i_truncate_pending == 0) { 1909 dout("__do_pending_vmtruncate %p none pending\n", inode); 1910 spin_unlock(&ci->i_ceph_lock); 1911 mutex_unlock(&ci->i_truncate_mutex); 1912 return; 1913 } 1914 1915 /* 1916 * make sure any dirty snapped pages are flushed before we 1917 * possibly truncate them.. so write AND block! 1918 */ 1919 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) { 1920 spin_unlock(&ci->i_ceph_lock); 1921 dout("__do_pending_vmtruncate %p flushing snaps first\n", 1922 inode); 1923 filemap_write_and_wait_range(&inode->i_data, 0, 1924 inode->i_sb->s_maxbytes); 1925 goto retry; 1926 } 1927 1928 /* there should be no reader or writer */ 1929 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref); 1930 1931 to = ci->i_truncate_size; 1932 wrbuffer_refs = ci->i_wrbuffer_ref; 1933 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode, 1934 ci->i_truncate_pending, to); 1935 spin_unlock(&ci->i_ceph_lock); 1936 1937 truncate_pagecache(inode, to); 1938 1939 spin_lock(&ci->i_ceph_lock); 1940 if (to == ci->i_truncate_size) { 1941 ci->i_truncate_pending = 0; 1942 finish = 1; 1943 } 1944 spin_unlock(&ci->i_ceph_lock); 1945 if (!finish) 1946 goto retry; 1947 1948 mutex_unlock(&ci->i_truncate_mutex); 1949 1950 if (wrbuffer_refs == 0) 1951 ceph_check_caps(ci, 0, NULL); 1952 1953 wake_up_all(&ci->i_cap_wq); 1954 } 1955 1956 static void ceph_inode_work(struct work_struct *work) 1957 { 1958 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info, 1959 i_work); 1960 struct inode *inode = &ci->vfs_inode; 1961 1962 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) { 1963 dout("writeback %p\n", inode); 1964 filemap_fdatawrite(&inode->i_data); 1965 } 1966 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask)) 1967 ceph_do_invalidate_pages(inode); 1968 1969 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask)) 1970 __ceph_do_pending_vmtruncate(inode); 1971 1972 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask)) 1973 ceph_check_caps(ci, 0, NULL); 1974 1975 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask)) 1976 ceph_flush_snaps(ci, NULL); 1977 1978 iput(inode); 1979 } 1980 1981 /* 1982 * symlinks 1983 */ 1984 static const struct inode_operations ceph_symlink_iops = { 1985 .get_link = simple_get_link, 1986 .setattr = ceph_setattr, 1987 .getattr = ceph_getattr, 1988 .listxattr = ceph_listxattr, 1989 }; 1990 1991 int __ceph_setattr(struct inode *inode, struct iattr *attr) 1992 { 1993 struct ceph_inode_info *ci = ceph_inode(inode); 1994 unsigned int ia_valid = attr->ia_valid; 1995 struct ceph_mds_request *req; 1996 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; 1997 struct ceph_cap_flush *prealloc_cf; 1998 int issued; 1999 int release = 0, dirtied = 0; 2000 int mask = 0; 2001 int err = 0; 2002 int inode_dirty_flags = 0; 2003 bool lock_snap_rwsem = false; 2004 2005 prealloc_cf = ceph_alloc_cap_flush(); 2006 if (!prealloc_cf) 2007 return -ENOMEM; 2008 2009 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR, 2010 USE_AUTH_MDS); 2011 if (IS_ERR(req)) { 2012 ceph_free_cap_flush(prealloc_cf); 2013 return PTR_ERR(req); 2014 } 2015 2016 spin_lock(&ci->i_ceph_lock); 2017 issued = __ceph_caps_issued(ci, NULL); 2018 2019 if (!ci->i_head_snapc && 2020 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) { 2021 lock_snap_rwsem = true; 2022 if (!down_read_trylock(&mdsc->snap_rwsem)) { 2023 spin_unlock(&ci->i_ceph_lock); 2024 down_read(&mdsc->snap_rwsem); 2025 spin_lock(&ci->i_ceph_lock); 2026 issued = __ceph_caps_issued(ci, NULL); 2027 } 2028 } 2029 2030 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued)); 2031 2032 if (ia_valid & ATTR_UID) { 2033 dout("setattr %p uid %d -> %d\n", inode, 2034 from_kuid(&init_user_ns, inode->i_uid), 2035 from_kuid(&init_user_ns, attr->ia_uid)); 2036 if (issued & CEPH_CAP_AUTH_EXCL) { 2037 inode->i_uid = attr->ia_uid; 2038 dirtied |= CEPH_CAP_AUTH_EXCL; 2039 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2040 !uid_eq(attr->ia_uid, inode->i_uid)) { 2041 req->r_args.setattr.uid = cpu_to_le32( 2042 from_kuid(&init_user_ns, attr->ia_uid)); 2043 mask |= CEPH_SETATTR_UID; 2044 release |= CEPH_CAP_AUTH_SHARED; 2045 } 2046 } 2047 if (ia_valid & ATTR_GID) { 2048 dout("setattr %p gid %d -> %d\n", inode, 2049 from_kgid(&init_user_ns, inode->i_gid), 2050 from_kgid(&init_user_ns, attr->ia_gid)); 2051 if (issued & CEPH_CAP_AUTH_EXCL) { 2052 inode->i_gid = attr->ia_gid; 2053 dirtied |= CEPH_CAP_AUTH_EXCL; 2054 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2055 !gid_eq(attr->ia_gid, inode->i_gid)) { 2056 req->r_args.setattr.gid = cpu_to_le32( 2057 from_kgid(&init_user_ns, attr->ia_gid)); 2058 mask |= CEPH_SETATTR_GID; 2059 release |= CEPH_CAP_AUTH_SHARED; 2060 } 2061 } 2062 if (ia_valid & ATTR_MODE) { 2063 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode, 2064 attr->ia_mode); 2065 if (issued & CEPH_CAP_AUTH_EXCL) { 2066 inode->i_mode = attr->ia_mode; 2067 dirtied |= CEPH_CAP_AUTH_EXCL; 2068 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 || 2069 attr->ia_mode != inode->i_mode) { 2070 inode->i_mode = attr->ia_mode; 2071 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode); 2072 mask |= CEPH_SETATTR_MODE; 2073 release |= CEPH_CAP_AUTH_SHARED; 2074 } 2075 } 2076 2077 if (ia_valid & ATTR_ATIME) { 2078 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode, 2079 inode->i_atime.tv_sec, inode->i_atime.tv_nsec, 2080 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec); 2081 if (issued & CEPH_CAP_FILE_EXCL) { 2082 ci->i_time_warp_seq++; 2083 inode->i_atime = attr->ia_atime; 2084 dirtied |= CEPH_CAP_FILE_EXCL; 2085 } else if ((issued & CEPH_CAP_FILE_WR) && 2086 timespec64_compare(&inode->i_atime, 2087 &attr->ia_atime) < 0) { 2088 inode->i_atime = attr->ia_atime; 2089 dirtied |= CEPH_CAP_FILE_WR; 2090 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2091 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) { 2092 ceph_encode_timespec64(&req->r_args.setattr.atime, 2093 &attr->ia_atime); 2094 mask |= CEPH_SETATTR_ATIME; 2095 release |= CEPH_CAP_FILE_SHARED | 2096 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2097 } 2098 } 2099 if (ia_valid & ATTR_SIZE) { 2100 loff_t isize = i_size_read(inode); 2101 2102 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size); 2103 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size > isize) { 2104 i_size_write(inode, attr->ia_size); 2105 inode->i_blocks = calc_inode_blocks(attr->ia_size); 2106 ci->i_reported_size = attr->ia_size; 2107 dirtied |= CEPH_CAP_FILE_EXCL; 2108 ia_valid |= ATTR_MTIME; 2109 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2110 attr->ia_size != isize) { 2111 req->r_args.setattr.size = cpu_to_le64(attr->ia_size); 2112 req->r_args.setattr.old_size = cpu_to_le64(isize); 2113 mask |= CEPH_SETATTR_SIZE; 2114 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL | 2115 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2116 } 2117 } 2118 if (ia_valid & ATTR_MTIME) { 2119 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode, 2120 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec, 2121 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec); 2122 if (issued & CEPH_CAP_FILE_EXCL) { 2123 ci->i_time_warp_seq++; 2124 inode->i_mtime = attr->ia_mtime; 2125 dirtied |= CEPH_CAP_FILE_EXCL; 2126 } else if ((issued & CEPH_CAP_FILE_WR) && 2127 timespec64_compare(&inode->i_mtime, 2128 &attr->ia_mtime) < 0) { 2129 inode->i_mtime = attr->ia_mtime; 2130 dirtied |= CEPH_CAP_FILE_WR; 2131 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 || 2132 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) { 2133 ceph_encode_timespec64(&req->r_args.setattr.mtime, 2134 &attr->ia_mtime); 2135 mask |= CEPH_SETATTR_MTIME; 2136 release |= CEPH_CAP_FILE_SHARED | 2137 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR; 2138 } 2139 } 2140 2141 /* these do nothing */ 2142 if (ia_valid & ATTR_CTIME) { 2143 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME| 2144 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0; 2145 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode, 2146 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec, 2147 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec, 2148 only ? "ctime only" : "ignored"); 2149 if (only) { 2150 /* 2151 * if kernel wants to dirty ctime but nothing else, 2152 * we need to choose a cap to dirty under, or do 2153 * a almost-no-op setattr 2154 */ 2155 if (issued & CEPH_CAP_AUTH_EXCL) 2156 dirtied |= CEPH_CAP_AUTH_EXCL; 2157 else if (issued & CEPH_CAP_FILE_EXCL) 2158 dirtied |= CEPH_CAP_FILE_EXCL; 2159 else if (issued & CEPH_CAP_XATTR_EXCL) 2160 dirtied |= CEPH_CAP_XATTR_EXCL; 2161 else 2162 mask |= CEPH_SETATTR_CTIME; 2163 } 2164 } 2165 if (ia_valid & ATTR_FILE) 2166 dout("setattr %p ATTR_FILE ... hrm!\n", inode); 2167 2168 if (dirtied) { 2169 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied, 2170 &prealloc_cf); 2171 inode->i_ctime = attr->ia_ctime; 2172 } 2173 2174 release &= issued; 2175 spin_unlock(&ci->i_ceph_lock); 2176 if (lock_snap_rwsem) 2177 up_read(&mdsc->snap_rwsem); 2178 2179 if (inode_dirty_flags) 2180 __mark_inode_dirty(inode, inode_dirty_flags); 2181 2182 2183 if (mask) { 2184 req->r_inode = inode; 2185 ihold(inode); 2186 req->r_inode_drop = release; 2187 req->r_args.setattr.mask = cpu_to_le32(mask); 2188 req->r_num_caps = 1; 2189 req->r_stamp = attr->ia_ctime; 2190 err = ceph_mdsc_do_request(mdsc, NULL, req); 2191 } 2192 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err, 2193 ceph_cap_string(dirtied), mask); 2194 2195 ceph_mdsc_put_request(req); 2196 ceph_free_cap_flush(prealloc_cf); 2197 2198 if (err >= 0 && (mask & CEPH_SETATTR_SIZE)) 2199 __ceph_do_pending_vmtruncate(inode); 2200 2201 return err; 2202 } 2203 2204 /* 2205 * setattr 2206 */ 2207 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry, 2208 struct iattr *attr) 2209 { 2210 struct inode *inode = d_inode(dentry); 2211 struct ceph_fs_client *fsc = ceph_inode_to_client(inode); 2212 int err; 2213 2214 if (ceph_snap(inode) != CEPH_NOSNAP) 2215 return -EROFS; 2216 2217 err = setattr_prepare(&init_user_ns, dentry, attr); 2218 if (err != 0) 2219 return err; 2220 2221 if ((attr->ia_valid & ATTR_SIZE) && 2222 attr->ia_size > max(i_size_read(inode), fsc->max_file_size)) 2223 return -EFBIG; 2224 2225 if ((attr->ia_valid & ATTR_SIZE) && 2226 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size)) 2227 return -EDQUOT; 2228 2229 err = __ceph_setattr(inode, attr); 2230 2231 if (err >= 0 && (attr->ia_valid & ATTR_MODE)) 2232 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode); 2233 2234 return err; 2235 } 2236 2237 /* 2238 * Verify that we have a lease on the given mask. If not, 2239 * do a getattr against an mds. 2240 */ 2241 int __ceph_do_getattr(struct inode *inode, struct page *locked_page, 2242 int mask, bool force) 2243 { 2244 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); 2245 struct ceph_mds_client *mdsc = fsc->mdsc; 2246 struct ceph_mds_request *req; 2247 int mode; 2248 int err; 2249 2250 if (ceph_snap(inode) == CEPH_SNAPDIR) { 2251 dout("do_getattr inode %p SNAPDIR\n", inode); 2252 return 0; 2253 } 2254 2255 dout("do_getattr inode %p mask %s mode 0%o\n", 2256 inode, ceph_cap_string(mask), inode->i_mode); 2257 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1)) 2258 return 0; 2259 2260 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS; 2261 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode); 2262 if (IS_ERR(req)) 2263 return PTR_ERR(req); 2264 req->r_inode = inode; 2265 ihold(inode); 2266 req->r_num_caps = 1; 2267 req->r_args.getattr.mask = cpu_to_le32(mask); 2268 req->r_locked_page = locked_page; 2269 err = ceph_mdsc_do_request(mdsc, NULL, req); 2270 if (locked_page && err == 0) { 2271 u64 inline_version = req->r_reply_info.targeti.inline_version; 2272 if (inline_version == 0) { 2273 /* the reply is supposed to contain inline data */ 2274 err = -EINVAL; 2275 } else if (inline_version == CEPH_INLINE_NONE) { 2276 err = -ENODATA; 2277 } else { 2278 err = req->r_reply_info.targeti.inline_len; 2279 } 2280 } 2281 ceph_mdsc_put_request(req); 2282 dout("do_getattr result=%d\n", err); 2283 return err; 2284 } 2285 2286 2287 /* 2288 * Check inode permissions. We verify we have a valid value for 2289 * the AUTH cap, then call the generic handler. 2290 */ 2291 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode, 2292 int mask) 2293 { 2294 int err; 2295 2296 if (mask & MAY_NOT_BLOCK) 2297 return -ECHILD; 2298 2299 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false); 2300 2301 if (!err) 2302 err = generic_permission(&init_user_ns, inode, mask); 2303 return err; 2304 } 2305 2306 /* Craft a mask of needed caps given a set of requested statx attrs. */ 2307 static int statx_to_caps(u32 want, umode_t mode) 2308 { 2309 int mask = 0; 2310 2311 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME)) 2312 mask |= CEPH_CAP_AUTH_SHARED; 2313 2314 if (want & (STATX_NLINK|STATX_CTIME)) { 2315 /* 2316 * The link count for directories depends on inode->i_subdirs, 2317 * and that is only updated when Fs caps are held. 2318 */ 2319 if (S_ISDIR(mode)) 2320 mask |= CEPH_CAP_FILE_SHARED; 2321 else 2322 mask |= CEPH_CAP_LINK_SHARED; 2323 } 2324 2325 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE| 2326 STATX_BLOCKS)) 2327 mask |= CEPH_CAP_FILE_SHARED; 2328 2329 if (want & (STATX_CTIME)) 2330 mask |= CEPH_CAP_XATTR_SHARED; 2331 2332 return mask; 2333 } 2334 2335 /* 2336 * Get all the attributes. If we have sufficient caps for the requested attrs, 2337 * then we can avoid talking to the MDS at all. 2338 */ 2339 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path, 2340 struct kstat *stat, u32 request_mask, unsigned int flags) 2341 { 2342 struct inode *inode = d_inode(path->dentry); 2343 struct ceph_inode_info *ci = ceph_inode(inode); 2344 u32 valid_mask = STATX_BASIC_STATS; 2345 int err = 0; 2346 2347 /* Skip the getattr altogether if we're asked not to sync */ 2348 if (!(flags & AT_STATX_DONT_SYNC)) { 2349 err = ceph_do_getattr(inode, 2350 statx_to_caps(request_mask, inode->i_mode), 2351 flags & AT_STATX_FORCE_SYNC); 2352 if (err) 2353 return err; 2354 } 2355 2356 generic_fillattr(&init_user_ns, inode, stat); 2357 stat->ino = ceph_present_inode(inode); 2358 2359 /* 2360 * btime on newly-allocated inodes is 0, so if this is still set to 2361 * that, then assume that it's not valid. 2362 */ 2363 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) { 2364 stat->btime = ci->i_btime; 2365 valid_mask |= STATX_BTIME; 2366 } 2367 2368 if (ceph_snap(inode) == CEPH_NOSNAP) 2369 stat->dev = inode->i_sb->s_dev; 2370 else 2371 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0; 2372 2373 if (S_ISDIR(inode->i_mode)) { 2374 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), 2375 RBYTES)) 2376 stat->size = ci->i_rbytes; 2377 else 2378 stat->size = ci->i_files + ci->i_subdirs; 2379 stat->blocks = 0; 2380 stat->blksize = 65536; 2381 /* 2382 * Some applications rely on the number of st_nlink 2383 * value on directories to be either 0 (if unlinked) 2384 * or 2 + number of subdirectories. 2385 */ 2386 if (stat->nlink == 1) 2387 /* '.' + '..' + subdirs */ 2388 stat->nlink = 1 + 1 + ci->i_subdirs; 2389 } 2390 2391 stat->result_mask = request_mask & valid_mask; 2392 return err; 2393 } 2394